Rodrigo de Salvo Braz
CurriculumVitae
Personal information
Birthday: / January 25, 1971, São Paulo, BrazilCitizenship: / Brazilian, US Permanent Resident
Mailing address: / 1390 Market St., apt 2316
San Francisco CA 94102
E-mail: / ,
Homepage: /
Home phone: / 415-519-0754
Current position
Senior Computer Scientist. Since September 2008.
Artificial Intelligence Center, SRI International.
Previous position
Post-doctoral researcher. From August 2007 to September 2008.
University of California, Berkeley (under Prof. Stuart Russell)
Education
Ph.D., Computer Science, October 2007.
Department of Computer Science, University of Illinois at Urbana-Champaign.
Thesis: Lifted First-Order Probabilistic Inference,
Advisor: Prof. Dan Roth.
Graduate student from September 1998 to May 2000
Department of Cognitive and Linguistic Sciences, Brown University.
Advisor: Prof. James Anderson.
M.Sc., Computer Science, January 1998.
Instituto de Matemática e Estatística da Universidade de São Paulo.
Thesis: "High level problems in neural networks" (Portuguese)
Advisor: Prof. Flávio Soares Corrêa da Silva.
B.Sc., Computer Science, December 1993.
Instituto de Matemática e Estatística da Universidade de São Paulo.
Research expertise
First-Order Probabilistic Inference (FOPI). This consists of my recent work with Dynamic Bayesian Logic (DBLOG) and my thesis subject, Lifted FOPI:
DynamicBayesian Logic (DBLOG): BLOG (by Milch and Russell) is an expressive language and semantics for probabilistic logic, including open universe assumptions. Current inference algorithms perform sampling while instantiating only the necessary random variables (there may be an infinite number of them). DBLOG extends BLOG to work with temporal models, in the same way Dynamic Bayesian networks extend Bayesian networks. We have developed a particle filter for DBLOG, as well as addressed several improvements on BLOG made more necessary by its use in DBLOG.
Lifted First-Order Probabilistic Inference. Most probabilistic models today are specified at a propositional level that renders them awkwardly limited in their expressivity. Some first-order probabilistic languages have been developed but their inference remains essentially propositional. This work develops a richer first-order (relational) language that allows more elegant specifications of knowledge and, more importantly, with an associated inference algorithm that is also at the first-order level. This makes it much faster than propositional inference approaches. An implementation is available. Possible applications are in Commonsense reasoning, Natural Language Processing, Human Computer Interaction and Vision, and many others.
Knowledge Representation in Natural Language Processing, applied in Kindle, an ARDA project on text entailment. This project is about representing several levels of sentences, from lexical to semantic, in a unified graph representation that is manipulated by inference rules. This way, one can detect whether two sentences can be equivalent or imply each other, even if they are very different superficially.
Multimodal Human-Computer Interface: I have been responsible for the decision module of the ITR Multimodal Tutoring project at the UIUC Beckman Institute. This interdisciplinary project, involving research groups in education, speech recognition, facial recognition, vision and reasoning, aimed at creating a multimodal computer tutoring environment for children to interact and learn about physical objects (in this case, Lego gears).
Machine Learning (Inductive Logic Programming, structured classification learning, Bayesian networks, decision trees, rule learning systems, neural networks).
Machine Learning algorithms implementation. Developed an object oriented C++ neural nets library, later extended to other C++ machine learning template algorithms, featuring several algorithms and data structures, for research.
Publications
Ph.D. Thesis
Lifted First-Order Probabilistic Inference.(improved (pdf), deposited (pdf), bibtex)
Ph.D. Thesis.
University of Illinois at Urbana-Champaign, 2007.
Related to FOPI.
Book Chapters
A Survey of First-Order Probabilistic Models. (pdf, bibtex)
D.E. Holmes and L.C. Jain (Eds.)
Innovations in Bayesian Networks, Studies in Computational Intelligence vol. 156, 2008.
with Eyal Amirand Dan Roth.
Related to FOPI.
Lifted First-Order Probabilistic Inference. (pdf, bibtex)
In L. Getoor and B. Taskar, (Eds.)
Introduction to Statistical Relational Learning. 2007.
with Eyal Amirand Dan Roth.
Related to FOPI.
An Inference Model for Semantic Entailment in Natural Language. (pdf, bibtex)
Lecture Notes in Computer Science vol. 3944, 2006.
Machine Learning Challenges, Evaluating Predictive Uncertainty, Visual Object Classification and Recognizing Textual Entailment, First PASCAL Machine Learning Challenges Workshop, Revised Selected Papers.
with Roxana Girju, Vasin Punyakanok, Dan Roth and Mark Sammons.
Related to Kindle.
Conference papers
Exact Inference for Relational Graphical Models with Interpreted Functions: Lifted Probabilistic Inference Modulo Theories. (revised version including supplementary material pdf,original pdf, original supplementary materials pdf, bibtex)
UAI-17: Conference on Uncertainty in Artificial Intelligence.
with Ciaran O’Reilly.
Related to FOPI.
Probabilistic Inference Modulo Theories. (revised pdf,original pdf, bibtex)
IJCAI-16: Twenty-Fifth International Joint Conference on Artificial Intelligence.
with Ciaran O’Reilly, Vibhav Gogate, and Rina Dechter.
Related to FOPI.
Exact lifted inference with distinct soft evidence on every object. (pdf, slides, bibtex)
AAAI-12: Twenty-Sixth Conference on Artificial Intelligence.
Hung Bui, Tuyen N. Huynh, Rodrigo de Salvo Braz.
Related to FOPI.
Efficient Methods for Lifted Inference with Aggregate Factors. (pdf, bibtex)
AAAI-11: Twenty-Fifth Conference on Artificial Intelligence.
Jaesik Choi, Rodrigo de Salvo Braz and Hung Bui.
Related to FOPI.
Gibbs Sampling in Open-Universe Stochastic Languages. (pdf, bibtex, ppt)
UAI-10: Uncertainty in Artificial Intelligence.
Nimar S. Arora, Rodrigo de Salvo Braz, Erik B. Sudderth, and Stuart Russell.
Related to BLOG.
MPE and Partial Inversion in Lifted Probabilistic Variable Elimination. (ps, pdf, bibtex)
AAAI-06: Twenty-First Conference on Artificial Intelligence.
with Eyal Amir, Dan Roth.
Related to FOPI.
Lifted First-Order Probabilistic Inference. (ps, pdf, bibtex)
IJCAI-05: Nineteenth International Joint Conference on Artificial Intelligence.
with Eyal Amir and Dan Roth.
Related to FOPI.
An Inference Model for Semantic Entailment in Natural Language. (ps, pdf, bibtex)
AAAI-05: Twentieth Conference on Artificial Intelligence.
with Roxana Girju, Vasin Punyakanok, Dan Roth, Mark Sammons.
Related to Kindle.
Workshops
Anytime Exact BeliefPropagation. (pdf,bibtex)
Seventh Statistical Relational AI (StaRAI-12) workshop at UAI'17
Gabriel Azevedo Ferreira, Charles Maussion, and Quentin Bertrand, Rodrigo de Salvo Braz.
Related to FOPI.
Probabilistic Inference Modulo Theories. (pdf,bibtex)
Workshop on Hybrid Reasoning at IJCAI 2015.
with Ciaran O’Reilly, Vibhav Gogate, and Rina Dechter.
Related to FOPI.
Lifted Arbitrary Constraint Solving for Lifted Probabilistic Inference. (pdf,bibtex)
2nd Statistical Relational AI (StaRAI-12) workshop at UAI'12
with Shahin Saadati, Hung Bui and Ciaran O’Reilly.
Related to FOPI.
Anytime Lifted Inference. (pdf, bibtex)
Oral Presentation
Statistical Relational Learning 2009
with Sriraam Natarajan, Hung Bui, Jude Shavlik and Stuart Russell.
Related to FOPI.
Dynamic BLOG.(ppt)
Poster presentation
Probabilistic Programming Languages Workshop
in NIPS 2008: 22nd Annual Conference onNeural Information Processing Systems
with Nimar Arora, Erik Sudderth and Stuart Russell.
Related to DBLOG.
MPE and Partial Inversion in Lifted Probabilistic Variable Elimination. (pdf, bibtex)
Statistical Relational Learning Workshop
in ICML 2006: 23rd International Conference on Machine Learning.
with Eyal Amir and Dan Roth.
Related to FOPI.
Knowledge Representation for Semantic Entailment and Question-Answering. (pdf, bibtex)
Knowledge and Reasoning for Question Answering Workshop
in IJCAI-05: Nineteenth International Joint Conference on Artificial Intelligence.
with Roxana Girju, Vasin Punyakanok, Dan Roth, Mark Sammons.
Related to Kindle.
Feature Extraction with Functional Subsumption in Description Logics.(ppt)
with Dan Roth.
Feature Extraction Workshop
in NIPS 2003: 17th Annual Conference onNeural Information Processing Systems.
Reading comprehension programs in a statistical-language-processing class. (ps, bibtex)
with Eugene Charniak, Yasemin Altun, Benjamin Garrett, Margaret Kosmala, Tomer Moscovich, Lixin Pang, Changhee Pyo, Ye Sun, Wei Wy, Zhongfa Yang, Shawn Zeller, and Lisa Zorn.
Reading Comprehension Tests as Evaluation for Computer-Based Language Understanding Systems Workshop in ANLP-NAACL 2000.
Miscellaneous
An Inference Model for Semantic Entailment and Question-Answering. (pdf)
Poster presentation at IJCAI 2005.
with Roxana Girju, Vasin Punyakanok, Dan Roth, Mark Sammons.
Related to Kindle.
Functional Subsumption in Feature Description Logic. (ps, bibtex)
Rodrigo de Salvo Braz and Dan Roth.
Technical Report UIUCDCS-R-2004-2423 (Engr. No UILU-ENG-2004-1724), UIUC Computer Science Department, April 2004.
Cross training and its application to skill mining. (pdf, bibtex)
IBM Systems Journal 41(3): 449-460 (2002).
Daniel Oblinger, Mark Reid, Mark Brodie, Rodrigo de Salvo Braz.
Integrating multiple domains and learning phases in a connectionist model.
Poster at Second Annual Graduate Conference of the Northeastern Cognitive Science Society (NECCS), April 1999.
Master Thesis
High level problems in neural networks. (doc(Portuguese), bibtex)
Master Thesis.
Universidade de São Paulo, 1998.
Professional experience
Postdoctoral Researcher for Prof. Stuart Russell at the University of California, Berkeley.
August 2007 – September 2008
Conducting research on Dynamic BLOG, a temporal version of BLOG (Bayesian LOGic) and on lifted first-order probabilistic inference on richer languages, as well as investigating bounded approximations for it.
Research Assistant for Prof. Dan Roth at the University of Illinois at Urbana-Champaign.
Fall 2006 – Summer 2007
Conducting research on a probabilistic version of logic resolution and Lifted First-Order Probabilistic Inference.
Research Assistant for Prof. Eyal Amir at the University of Illinois at Urbana-Champaign.
Summer 2006
Conducting research on a probabilistic version of logic resolution.
Summer intern, Cycorp.
Summer 2005.
Furthering Lifted First-Order Probabilistic Inference research toward applying it to the Cyc project.
Research Assistant for Prof. Dan Roth at the University of Illinois at Urbana-Champaign.
Fall 2002 – Spring 2006
Building a ITR Multimodal Tutoring application with Lifted First-Order Probabilistic Inference. The goal is to have an application to use facial, speech and image recognition to guide a student through a session learning and playing with physical objects (Lego gears).
Summer intern, NCSA – National Center for Supercomputing Applications.
Summer 2002.
Working on an probabilistic relational learning algorithm to be incorporated into NCSA's D2K learning framework.
Summer intern, Thomas J. Watson Research Lab, IBM.
Summer 2001.
Working on Machine Learning project for automatic employee skill assessment from email under supervision of Dr. Dan Oblinger and Dr. Robert Farrell. Co-author in related paper, “Cross Training and Its Application to Skill Mining.”
Systems Analyst, UOL Internacional, Argentina.
Summer 2000.
Writing performance assessment software for the largest Internet access and content provider in Latin America, dealing with data from its branches in seven countries.
Systems Analyst, Insite Internet Solutions, Brazil
December 1997 to August 1998.
Programming of Internet software in Perl, Cold Fusion and others. Among other things, developed an online incremental log analyzer that runs as a daemon. The user defines operators to be applied continually to each log under its monitoring. Data are always updated and available.
Systems Analyst, Back-up Systems, Brazil.
June 1996 to November 1997.
Systems programming and support. Developed several database systems using C++ and a ZIP code finder robust to spelling errors and incomplete information.
Software Analyst and Associate Editor, PC Magazine Brazil.
May 1994 to May 1995.
Analysis and writing on commercial software. Member of a three-person team in charge of design and execution of independent tests aimed to assess performance and quality of software from all major vendors of a given area. Each issue was dedicated to a different kind of software. The team was also responsible for writing the articles to be published on the magazine, the major computer magazine in the country at the time.
Systems Analyst, Bull Systems, Brazil,
August 1993 to May 1994.
Analysis and programming using artificial intelligence tools (mostly Charme, a constraint programming language which was the precursor to ILOG Solver). Bull Systems was the Brazilian branch of the French Bull. The main application was on scheduling sugar-cane transportation involving some 500 farms.
Development of critical business database systems (in Borland C++ Builder / Interbase) for a nationwide company in Brazil, Saúde e Vida, which has been running since 1998.
Programming Expertise
Highly proficient programmer in Java, C++, and Perl.
Experienced in the Borland C++ Builder framework, especially database programming.
Web and database programming in Perl.
Teaching experience
Teaching Assistant for Professor Dan Roth, UIUC, Spring 2002.
"Pattern Recognition and Machine Learning".
Teaching Assistant for Professor David Kriegman, UIUC, Fall 2001.
"Introduction to Artificial Intelligence".
Teaching Assistant for Professor Gerald DeJong, UIUC, Spring 2001.
"Introduction to Artificial Intelligence".
Teaching Assistant for Professor James Anderson, Brown University, Spring 2000.
"Neural Modeling Laboratory".
Teaching Assistant for Professor Steven Sloman, Brown University, Fall 1999.
"Making Decisions".
Mini course on C++ for graduate students, Brown University, June 1999.
Awards
IGERT fellowship, NSF, Spring-Summer 1999, Brown University.
CNPq fellowship, 1996-1997, Brazilian government.
Outstanding Graduate Student Service Award, 2000, Dept. of Computer Science, UIUC.
Service
Journal reviewer
Artificial Intelligence Journal, Elsevier, 2014, 2015.
Journal of Artificial Intelligence Research (JAIR)
Mathematical Structures in Computer Science Special Issue on WoLLIC 2015
Area Chair
IJCAI 2015 (International Joint Conference in Artificial Intelligence)
Workshops co-organizer
SymInfOpt-17: AAAI-17 Workshop on Symbolic Inference and Optimization
DeLBP 2016: AAAI-16 Workshop on Declarative Learning Based Programming
Senior Program Committee member
IJCAI 2013 (International Joint Conference in Artificial Intelligence)
Program Committee member
IJCAI 2017,2011,2009 (International Joint Conference in Artificial Intelligence)
DeLBP 2017 (2nd Int’l Workshop on Declarative Learning Based Programming)
PGM 2016 (International Conference in Probabilistic Graphical Models)
IJCAI 2009, 2011,2017 (International Joint Conference in Artificial Intelligence)
AAAI 2014, 2013, 2007 (Assoc. for the Advancement of Artificial Intelligence)
UAI 2010, 2009 (Uncertainty in Artificial Intelligence)
SRL 2009 (Statistical Relational Learning)
AISTATS 2009 (Artificial Intelligence and Statistics)
IBERAMIA’08 (Ibero-American Conference on Artificial Intelligence)
SBIA’12, ’08 (Brazilian Symposium on Artificial Intelligence)
Research grant proposal evaluator
KU Leuven (Belgium), 2017
References
Professor Stuart Russell, UC Berkeley,
Professor Dan Roth, UIUC,
Professor Eyal Amir, UIUC,
Other Languages
Portuguese, some Italian, Spanish, and French.